On June 9, Anthropic released Claude Fable 5 — the first Mythos-class model available to the general public. The short version: capabilities above anything they have ever made generally available, state-of-the-art on nearly every benchmark that matters — software engineering, knowledge work, vision, scientific research.
The interesting part is the release design. Fable 5 ships with conservative safeguards: queries on certain sensitive topics get answered by Claude Opus 4.8 instead, which Anthropic says triggers in less than 5% of sessions. The unrestricted sibling, Claude Mythos 5 — same underlying model, safeguards lifted in some areas — goes only to a small group of cyberdefenders and infrastructure providers through Project Glasswing, in collaboration with the US government.
Why I care, as someone who ships with agents
My entire workflow runs on AI agents doing the heavy lifting across 11 products. Model jumps like this are not abstract benchmark news for me — they change what I can delegate. Better software-engineering performance means longer autonomous runs, fewer babysitting interventions, and more of the boring-but-critical work (migrations, test suites, refactors) done end-to-end.
The pricing is the other headline: $10 per million input tokens and $50 per million output — less than half of what Mythos Preview cost. Frontier capability getting cheaper faster than I can spend it is the single best trend in this business.
I'll be running Fable 5 through my own agent pipeline over the next weeks — if the arena numbers are interesting, they'll end up in an article here.